Object state evaluation method, device, computer equipment and system

Abstract

The application relates to a state evaluation method and device of an object, computer equipment and a system. The method comprises the following steps: acquiring a first crack image of an object to be evaluated; using a regression model to perform prediction based on a first crack parameter in the first crack image, so as to obtain a first stress intensity factor of the object to be evaluated; the regression model is a model for predicting the stress intensity factor, and the regression model is obtained by training a second crack parameter and a second stress intensity factor in a second crack image of a target object, and the second stress intensity factor is simulated based on the second crack parameter; and the state of the object to be evaluated is evaluated based on the first stress intensity factor, so as to obtain a state evaluation result of the object to be evaluated. The method can be used for state evaluation in multiple scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer vision and fracture mechanics analysis, and in particular to a state evaluation method and device for an object, a computer device and system. BACKGROUND

[0002] In various production activities, most objects are prone to be affected by fatigue load, corrosion and environment during long-term service, causing the generation and expansion of micro cracks. If these cracks are not identified and accurately evaluated in time, the service safety and reliable life of the object will be seriously threatened.

[0003] At present, when the existing technology detects cracks and analyzes stress, the entire crack propagation process is always assumed to be semi-elliptical, which makes it difficult to cover the high complexity of the crack morphology commonly seen in actual inspection. Moreover, the existing technology mainly relies on high-performance special equipment, which not only has high procurement and maintenance costs, but also requires professional on-site operation, making it difficult to be applied to crack inspection in high places, narrow holes or underwater. SUMMARY

[0004] Therefore, it is necessary to provide a state evaluation method and device for an object, a computer device and system capable of state evaluation in multiple scenarios to solve the above technical problems.

[0005] A state evaluation method for an object, the method comprising:

[0006] obtaining a first crack image of an object to be evaluated;

[0007] based on a first crack parameter in the first crack image, using a regression model for prediction to obtain a first stress intensity factor of the object to be evaluated; the regression model is a model for predicting the stress intensity factor, and the regression model is obtained by training a second crack parameter and a second stress intensity factor in a second crack image of a target object, and the second stress intensity factor is obtained by simulation from the second crack parameter;

[0008] based on the first stress intensity factor, evaluating the state of the object to be evaluated to obtain a state evaluation result of the object to be evaluated.

[0009] A state evaluation device for an object, the device comprising:

[0010] an image acquisition unit configured to acquire a first crack image of an object to be evaluated;

[0011] a prediction unit configured to predict, based on a first crack parameter in the first crack image, a first stress intensity factor of the to-be-evaluated object using a regression model, the regression model being a model for predicting a stress intensity factor, the regression model being trained by a second crack parameter in a second crack image of a target object and a second stress intensity factor simulated based on the second crack parameter;

[0012] a state evaluation unit configured to evaluate a state of the to-be-evaluated object based on the first stress intensity factor, and obtain a state evaluation result of the to-be-evaluated object.

[0013] A computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0014] A state evaluation system of an object includes a crack detection unit, a parameter conversion unit, a simulation unit, a training and prediction unit, and a control unit, and the system is characterized in that:

[0015] The crack detection unit is configured to detect cracks in a first crack image of a to-be-evaluated object and cracks in a second crack image of a target object.

[0016] The parameter conversion unit is configured to obtain coordinate information of the cracks in the second crack image and convert the coordinate information.

[0017] The simulation unit is configured to obtain a second stress intensity factor of the target object.

[0018] The training and prediction unit is configured to train a regression model and predict a stress intensity factor.

[0019] The control unit is configured to implement the steps of the above method.

[0020] The state evaluation method, device, computer device and system of the object, by acquiring the first crack image of the object to be evaluated, based on the first crack parameter in the first crack image, using the regression model for prediction, obtaining the first stress intensity factor of the object to be evaluated; wherein the regression model is a model for predicting the stress intensity factor, the regression model is obtained by training the second crack parameter and the second stress intensity factor in the second crack image of the target object, and the second stress intensity factor is simulated by the second crack parameter. In this way, when the second crack parameter in the second crack image is obtained, the second crack parameter can be directly simulated, the transmission process of the second crack parameter is simplified, the speed of obtaining the second stress intensity factor is improved, so that the regression model for predicting the stress intensity factor can be quickly obtained, the prediction efficiency of the first stress intensity factor is improved, and the acquisition efficiency of the state evaluation result of the object to be evaluated is improved. By using the regression model to predict based on the first crack parameter in the first crack image, the nonlinear estimation of the first stress intensity factor in various scenes can be realized only by the first crack parameter of the first crack image without on-site operation. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 An application environment diagram of the object state evaluation method in an embodiment;

[0022] Figure 2 A flowchart of the object state evaluation method in an embodiment;

[0023] Figure 3 A schematic diagram of the second crack image in an embodiment;

[0024] Figure 4 A schematic diagram of the second crack image in another embodiment;

[0025] Figure 5 A schematic diagram of the second crack image in another embodiment;

[0026] Figure 6 A schematic diagram of the second crack image in another embodiment;

[0027] Figure 7 A schematic diagram of the second crack image in another embodiment;

[0028] Figure 8 A schematic diagram of the segmentation effect of the CE-NET model on the second crack image in an embodiment; Figure 7

[0029] Figure 9 A schematic diagram of the segmentation effect of the lightweight crack segmentation network on the second crack image in an embodiment; Figure 7

[0030] ​​Figure 10 is a schematic diagram of a second crack image in another embodiment;

[0031] Figure 11 is a global feature and boundary feature heat map of the lightweight crack segmentation network for Figure 10 ;

[0032] Figure 12 is a global feature and boundary feature heat map of the SCRN model for Figure 10 ;

[0033] Figure 13 is a global feature and boundary feature heat map of the CE-Net model for Figure 10 ;

[0034] Figure 14 is a structural diagram of a lightweight crack segmentation network in an embodiment;

[0035] Figure 15 is a flowchart of a state evaluation method of an object in another embodiment;

[0036] Figure 16 is a structural block diagram of a state evaluation device of an object in an embodiment;

[0037] Figure 17 is an internal structure diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0038] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0039] The state evaluation method of an object provided by the embodiments of the present application can be applied to, for example, Figure 1The application environment shown. Among them, the control unit 102 interacts with the crack detection unit 104, the parameter conversion unit 106, the simulation unit 108 and the training prediction unit 110 through wired channels / wireless channels. The crack detection unit 104 is used to detect the cracks in the first crack image of the object to be evaluated, and detect the cracks in the second crack image of the target object; the parameter conversion unit 106 is used to obtain the coordinate information of the cracks in the second crack image, and convert the coordinate information; the simulation unit 108 is used to obtain the second stress intensity factor of the target object; the training prediction unit 110 is used to train the regression model and predict the stress intensity factor; the data storage system can store the data required by the control unit 102 for processing. The control unit 102 obtains the first crack image of the object to be evaluated; the control unit 102 uses the regression model to predict based on the first crack parameter in the first crack image to obtain the first stress intensity factor of the object to be evaluated; the regression model is a model for predicting the stress intensity factor, and the regression model is obtained by training the second crack parameter and the second stress intensity factor in the second crack image of the target object, and the second stress intensity factor is obtained by simulation from the second crack parameter; the control unit 102 evaluates the state of the object to be evaluated based on the first stress intensity factor to obtain the state evaluation result of the object to be evaluated.

[0040] In one embodiment, as shown in Figure 2 , a state evaluation method of an object is provided, and the method is applied to Figure 1 the control unit 102 as an example, including the following steps:

[0041] S202, obtaining the first crack image of the object to be evaluated.

[0042] Among them, the object to be evaluated refers to the object with cracks. The object to be evaluated includes but is not limited to various metal materials, ceramic materials, composite materials, biological materials, building materials and geological materials. Further, the metal material can be steel; the composite material can be carbon fiber reinforced plastic and glass fiber reinforced plastic; the biological material can be bone and soft tissue; the building material can be wood and concrete; the geological material can be rock.

[0043] The first crack image is an image collected from the crack of the object to be evaluated. Further, an industrial camera with a resolution reaching a first threshold is used to collect the first crack image at the crack of the object to be evaluated to obtain the first crack image of the object to be evaluated.

[0044] In some embodiments, the first crack image can be an image collected under different lighting conditions, or an image collected for an object to be evaluated with different surface textures and sizes. The cracks in the first crack image can have different lengths, widths, and color characteristics. The first crack image can be an image with marked information. The marked information can be one or more of handwritten marks, a scale, and color marks. Further, the lighting conditions include, but are not limited to, image brightness conditions, contrast conditions, and sharpness conditions.

[0045] In S204, based on the first crack parameter in the first crack image, a regression model is used for prediction to obtain the first stress intensity factor of the object to be evaluated. The regression model is a model for predicting stress intensity factors, and the regression model is obtained by training the second crack parameter and the second stress intensity factor in the second crack image of the target object. The second stress intensity factor is simulated from the second crack parameter.

[0046] The regression model is a statistical method for prediction and analysis, mainly used to understand the relationship between one or more independent variables and one dependent variable. The regression model includes, but is not limited to, linear regression, polynomial regression, ridge regression, lasso regression, elastic net regression, logistic regression, support vector regression, decision tree regression, random forest regression, gradient boosting regression, K-nearest neighbors, Huber regression, and partial least squares.

[0047] The first crack parameter is a parameter identified and extracted from the first crack image of the object to be evaluated. The first crack parameter includes, but is not limited to, crack length, size ratio, bifurcation angle, and stress ratio. The crack length refers to the straight-line distance between the two endpoints of the crack. The bifurcation angle refers to the included angle between the two crack branches. The size ratio refers to the ratio of the bifurcated crack length to the bifurcation angle. The stress ratio is the ratio of normal stress to shear stress. The first crack parameter can be obtained by manual acquisition or based on a segmentation model. Further, a lightweight crack segmentation network including an enhanced multi-scale attention fusion module and a multi-level edge detail enhancement module is established. The first coordinate information of the crack in the first crack image of the object to be evaluated is obtained using the lightweight crack segmentation network. When the conversion parameter is obtained, the conversion parameter is used to convert the first coordinate information to obtain the second coordinate information after conversion. Based on the second coordinate information, the first crack parameter in the first crack image is calculated.

[0048] The first stress intensity factor includes at least one of an opening stress intensity factor and a sliding stress intensity factor. The opening stress intensity factor is also referred to as a stress intensity factor in an opening mode or a tensile mode, which refers to the stress field intensity acting on the crack surface to make it open in a direction perpendicular to the crack surface. The sliding stress intensity factor is a stress intensity factor in a sliding mode or a shear mode, which describes the stress field intensity making the two sides of the crack slide in parallel but opposite directions.

[0049] The second crack parameter is a parameter identified and extracted from a second crack image of the target object. The target object refers to an object with cracks. The target object includes but is not limited to various metal materials, ceramic materials, composite materials, biological materials, building materials, and geological materials.

[0050] The second stress intensity factor corresponding to the second crack parameter can be obtained by simulation modeling through a finite element simulation software. Further, after obtaining the second crack parameter from the second crack image of the target object, the control unit directly inputs the obtained second crack parameter into the finite element simulation software, so as to realize seamless docking of the second crack parameter to the mechanical modeling, improve the generation efficiency of the simulation sample, and thus enable the second stress intensity factor to be used for model training more quickly, so as to quickly obtain the trained regression model.

[0051] The second crack image can be an image collected from a crack of the object to be evaluated, or an image obtained from an official database. In this way, the data source of the second crack image has dual nature, thereby improving the accuracy and reliability of the second crack image and enhancing the comprehensiveness of the second crack image and reducing bias. The number of second crack images is multiple frames, and the resolution of each frame of the second crack image can be consistent or inconsistent.

[0052] In some embodiments, the second crack image has a multi-modal characteristic, specifically represented as: the background environment in the second crack image has diversity, including different light conditions, surface textures, and sizes; the crack morphology in the second crack image has heterogeneity, covering cracks with different lengths, widths, and color characteristics; the second crack image has marking information, which can be one or more of handwritten marks, scales, and color marks. As shown in Figure 3 、 Figure 4 、 Figure 5 , the second crack image with marking information is as shown in Figure 6 .

[0053] S206, evaluating the state of the object to be evaluated based on the first stress intensity factor to obtain a state evaluation result of the object to be evaluated.

[0054] The state evaluation result of the to-be-evaluated object includes one or more of a health state and a service life of the to-be-evaluated object. The state of the to-be-evaluated object is evaluated to determine whether the to-be-evaluated object is safe in the presence of cracks.

[0055] In some embodiments, the state evaluation method of the object further includes determining a maintenance strategy for the to-be-evaluated object according to the state evaluation result of the to-be-evaluated object. Specifically, if the state evaluation result of the to-be-evaluated object indicates that the to-be-evaluated object is potentially dangerous, the monitoring frequency of the to-be-evaluated object needs to be increased, repair measures need to be implemented, or damaged components need to be replaced.

[0056] In the above state evaluation method of the object, the first crack image of the to-be-evaluated object is obtained, the first stress intensity factor of the to-be-evaluated object is obtained by using a regression model to predict based on the first crack parameter in the first crack image, the regression model is a model for predicting the stress intensity factor, and the regression model is obtained by training the second crack parameter and the second stress intensity factor in the second crack image of the target object, and the second stress intensity factor is obtained by simulation based on the second crack parameter. In this way, when the second crack parameter in the second crack image is obtained, the second crack parameter can be directly used for simulation, the transmission process of the second crack parameter is simplified, the speed of obtaining the second stress intensity factor is improved, and thus the regression model for predicting the stress intensity factor can be quickly obtained, the prediction efficiency of the first stress intensity factor is improved, and the efficiency of obtaining the state evaluation result of the to-be-evaluated object is improved. By using the regression model to predict based on the first crack parameter in the first crack image, the nonlinear estimation of the first stress intensity factor in various scenarios can be realized only by the first crack parameter of the first crack image without on-site operation.

[0057] In one embodiment, the state evaluation method of the object further includes:

[0058] Obtaining the second crack parameter in the second crack image of the target object.

[0059] Establishing a plurality of crack models based on the second crack parameter, and applying boundary conditions and loads to each crack model for simulation to obtain the second stress intensity factor under various working conditions.

[0060] Training the regression model based on the second crack parameter and the second stress intensity factor to obtain the regression model for predicting the stress intensity factor.

[0061] The second crack parameters in the second crack image can be obtained by a model or by manual operation. The second crack parameters include, but are not limited to, crack length, size ratio, bifurcation angle, and stress ratio. In some embodiments, if the target object and the object to be evaluated are steel materials, the range of the stress ratio σ / τ can be set to 0.2-10 with an increment of 0.01 according to the typical stress level of the steel materials, wherein the normal stress σ is fixed at 100 MPa, and the shear stress τ is calculated from the stress ratio.

[0062] The crack model is established by a finite element simulation software. Specifically, the control unit writes a script to drive the finite element simulation software, constructs a crack model in the finite element simulation software according to the input second crack parameters, applies boundary conditions and loads to the constructed crack model for simulation, and obtains a plurality of second stress intensity factors.

[0063] The boundary conditions applied to each crack model are consistent. The loads applied to each crack model are determined according to the stress ratio corresponding to the crack model.

[0064] The second stress intensity factor includes at least one of an opening-type stress intensity factor and a sliding-type stress intensity factor.

[0065] The number of trained regression models can be multiple or one. If the number of trained regression models is multiple, the regression model with the best effect is used to predict the stress intensity factor of the first crack image.

[0066] In some embodiments, the regression model is trained based on the second crack parameters and the second stress intensity factors, including: comparing the plurality of second stress intensity factors with a closed-form solution, and determining that the second stress intensity factor is reliable when the error is less than a second threshold; and training the regression model based on the second crack parameters and the reliable second stress intensity factor. The closed-form solution is a solution calculated using an explicit mathematical expression based on the second crack parameters. Further, if the error is greater than the second threshold, the parameters in the crack model are adjusted, or the boundary conditions and the loads are adjusted, until the error is less than the second threshold.

[0067] In this embodiment, the second crack parameters in the second crack image of the target object are obtained, and a plurality of crack models are established based on the second crack parameters. This can realize seamless connection of the second crack parameters to the mechanical modeling, simplify the transmission process of the second crack parameters, improve the efficiency of obtaining the second stress intensity factor, and thus improve the training efficiency of the regression model.

[0068] In one embodiment, obtaining the second crack parameters in the second crack image of the target object includes:

[0069] The coordinate information of the crack in the second crack image of the target object is obtained.

[0070] When the conversion parameter is obtained, the coordinate information is converted using the conversion parameter to obtain converted crack coordinates.

[0071] Based on the crack coordinates, the second crack parameter is determined.

[0072] The coordinate information is on-image information in the second crack image, and the on-image information is obtained after conversion from the electromagnetic space. Therefore, the coordinate information should also be converted into crack coordinates in the electromagnetic space, so that the regression model can be accurately trained using the second crack parameter. The coordinate information can be obtained by a segmentation model or manually.

[0073] The conversion parameter is data used to convert the coordinate information into crack coordinates. The conversion parameter can be scale information of the second crack image, or a perspective transformation matrix used to estimate the second crack image to the electromagnetic space. Specifically, the coordinate information is converted into crack coordinates by the perspective transformation matrix A, and the formula is:

[0074]

[0075] where P DLT (X, Y) is the converted crack coordinates, P Original (x, y) is the coordinate information in the second crack image before conversion.

[0076] In this embodiment, by obtaining the coordinate information of the crack in the second crack image of the target object, when the conversion parameter is obtained, the coordinate information is converted using the conversion parameter. This can avoid manual conversion of the coordinate information, thereby providing high-precision input parameters for constructing the crack model and training the regression model, improving the accuracy of the second stress intensity factor obtained by simulation, and thereby improving the prediction effect of the regression model.

[0077] In some embodiments, obtaining the second crack parameter in the second crack image of the target object includes: obtaining an initial crack parameter in the second crack image of the target object; when a calibration parameter is obtained, calibrating the initial crack parameter using the calibration parameter to obtain the second crack parameter; and when the calibration parameter for the initial crack parameter is not obtained, determining the calibration parameter based on shooting information used to shoot the second crack image.

[0078] The initial crack parameter is a crack parameter obtained according to on-image distances in the second crack image, and the on-image distances are obtained after a certain scale reduction. Therefore, the initial crack parameter should also be converted into the second crack parameter in the electromagnetic space, so that the regression model can be accurately trained using the second crack parameter. The initial crack parameter can be obtained by a model or manually.

[0079] The calibration parameter is data for converting the initial crack parameter to the second crack parameter. The calibration parameter can be scale information of the second crack image, or a perspective transformation matrix for estimating a distance on a map of the second crack image to an electromagnetic space.

[0080] The shooting information is an internal parameter and an external parameter of the camera, including but not limited to a focal length, an object distance, a lens characteristic, a sensor characteristic, and a shooting angle.

[0081] In an embodiment, the coordinate information of the crack in the second crack image of the target object is obtained, including:

[0082] A lightweight crack segmentation network is established. The lightweight crack segmentation network includes an enhanced multi-scale attention fusion module and a multi-level edge detail enhancement module.

[0083] The coordinate information of the crack in the second crack image of the target object is obtained using the lightweight crack segmentation network.

[0084] The enhanced multi-scale attention fusion module extracts different receptive field features through multi-scale convolution kernels and fuses them using channel and spatial attention, which can effectively enhance the expression of low-contrast crack morphology. The multi-level edge detail enhancement module combines gradient enhancement and parameter-free attention mechanism to enhance image details in edge regions, which can improve the recognition accuracy and continuity of crack boundaries.

[0085] When the coordinate information of the crack in the second crack image of the target object is obtained using the lightweight crack segmentation network, only a normal camera or a drone is needed for crack focusing shooting, without the need for rust removal or three-dimensional scanning equipment. This can significantly reduce hardware costs and improve deployment flexibility, adapting to various complex inspection scenarios. And through the enhanced multi-scale attention fusion module and the multi-level edge detail enhancement module, the micro-crack edge is finely extracted, even in the presence of various visual challenges such as texture interference, shadows, and rust, the complex morphology including bifurcation, connectivity, and jaggedness can be restored, providing a high-fidelity premise for subsequent geometric quantization.

[0086] Further, the lightweight crack segmentation network further includes an encoder part and a decoder part. The encoder part uses deep separable convolution combined with an efficient channel attention mechanism to extract multi-scale features, and suppresses channel redundancy through the efficient channel attention mechanism, focuses on defect sensitive areas, improves feature extraction capability, compresses redundant information, and enhances the response to small crack areas. The decoder part uses a hybrid upsampling and jump connection mechanism to fuse shallow edge information and deep semantic features, which can improve the segmentation integrity of small target cracks.

[0087] Furthermore, when using a lightweight crack segmentation network to obtain the coordinate information of the crack in the second crack image of the target object, the network outputs the crack region segmented by the lightweight crack segmentation network, and the coordinate information marked in the crack region is extracted. Further, if the marked coordinate information is not transformed, initial crack parameters can be calculated based on the coordinate information, and then transformed using calibration parameters to obtain the second crack parameters. For example, crack length, size ratio, and bifurcation angle can be calculated based on the coordinate information.

[0088] In this embodiment, a lightweight crack segmentation network is established, including an enhanced multi-scale attention fusion module and a multi-level edge detail enhancement module. This lightweight network is then used to obtain the coordinate information of the cracks in the second crack image of the target object. This enhances robustness and boundary restoration capabilities, enabling high-precision segmentation and boundary restoration of minute cracks even in complex backgrounds and low-contrast conditions. The network obtains accurate crack coordinate information from the second crack image, overcoming the problem of insufficient fine-grained feature representation in traditional models. Furthermore, the lightweight design and boundary enhancement mechanism of the lightweight crack segmentation network do not require additional large-kernel convolutions or Transformer structures, resulting in faster training and inference speeds and higher energy efficiency.

[0089] In some embodiments, lightweight crack segmentation networks can also be evaluated using one or more of the following metrics: performance metrics and efficiency metrics.

[0090] Regarding performance metrics, crack detection tasks primarily focus on the distribution of defects in the image. Therefore, the miIoU (mean image intersection-to-union ratio) and miDice (mean image dice) coefficients are introduced to determine the performance metrics. Specifically:

[0091] Let the prediction result of the nth image of the second crack be... The second crack image has a width of W pixels and a height of H pixels. In a binary classification problem, pixels are classified as positive or negative according to a set classification threshold; a commonly used threshold is 0.5.

[0092]

[0093] in, Let be the predicted value of position (i,j) in the nth image of the second crack; This represents the predicted classification result; the true label for the nth image of the second crack is G. n ∈{0,1} 1×H×W By comparison and G n You can obtain TP. n 、FN n and FP n , of which TPn FN represents the number of true negative pixels n FP represents the number of false positive pixels n mi IoU and mi Dice can be calculated by the following formula:

[0094]

[0095] where N is the total number of second crack images; ε is a constant with a small value, mainly used to prevent the denominator from being zero, usually set to 10 -6 .

[0096] In this embodiment, Mi IoU and mi Dice are used as complementary evaluation indicators, both of which are positively correlated with the performance of the segmentation model. Mi Dice is based on global similarity and emphasizes the structural consistency at the image level; while mi IoU measures the pixel-level overlap and highlights the geometric accuracy of local regions. The joint application of Mi IoU and mi Dice can effectively integrate the quantitative information of macro structure and micro details, and improve the accuracy and comprehensiveness of the evaluation of the lightweight crack segmentation model.

[0097] For efficiency indicators, network parameter quantity and product-sum operation quantity can be introduced to evaluate the computational efficiency of the lightweight crack segmentation network. Network parameter quantity reflects the complexity of the model structure, defined as the total number of trainable parameters; product-sum operation quantity represents the computational intensity of forward inference, which is used to accurately measure the floating-point operation cost.

[0098] In some embodiments, the lightweight crack segmentation network can also be compared with multiple segmentation networks. Specifically, Table 1 compares the performance of the lightweight crack segmentation network and other segmentation networks on the second crack image.

[0099] The results show that the lightweight crack segmentation network outperforms existing segmentation network models in the crack segmentation task with only 2.0M Param (parameter quantity) and 10.21G MACs (product-sum operation quantity) computational overhead, achieving significant advantages in mi IoU = 81.74% and mi Dice = 88.45% indicators, which reflects the excellent balance of high precision and low computational cost, suitable for efficient inference and training in resource-constrained environments. The performance improvement is due to the introduction of the enhanced multi-scale attention fusion module and the multi-level edge detail enhancement module in the lightweight crack segmentation network, which effectively refines the crack boundary and enhances the segmentation accuracy and noise resistance of the edge region. In addition, Figure 7 for the second crack image, Figure 8 for the segmentation effect of the CE-NET (Contextual Encoder-Net, contextual encoder network) model on Figure 7 the second crack image,Figure 9 For lightweight crack segmentation networks Figure 7 The segmentation results of the second crack image show that the crack image predicted by the lightweight crack segmentation network has clearer boundaries, less noise, and higher consistency with the real annotation. This further verifies the advantages of the lightweight crack segmentation network in terms of robustness and generalization ability.

[0100] Table 1 provides a quantitative evaluation of the results for different network models.

[0101]

[0102] To further analyze the decision-making mechanism of the model, heatmaps corresponding to multiple segmentation models were generated based on the Grad-CAM (gradient-weighted class activation mapping) method. Figure 10 Image of the second crack. Figure 11 For lightweight crack segmentation networks Figure 10 Global and boundary feature heatmaps Figure 12 For SCRN (Segmental Continuous Recurrent Network) models Figure 10 Global and boundary feature heatmaps Figure 13 For the CE-Net model Figure 10 The global and boundary feature heatmaps are shown, with the locations of artificially labeled and real cracks marked with black arrows. Heatmap analysis indicates that the lightweight crack segmentation network can accurately capture crack contours in low-contrast, complex backgrounds, and varied crack morphologies, and effectively suppresses misjudgments caused by artificial labels, validating the effectiveness of multi-scale feature fusion and boundary enhancement design. Therefore, the lightweight crack segmentation network exhibits stronger boundary clarity and noise suppression capabilities in various complex scenes. The heatmaps show that, compared to SCRN and CE-Net models, the lightweight crack segmentation network can still accurately focus on real crack regions in low-contrast cracks and complex texture backgrounds, and significantly suppress erroneous responses caused by pseudo-labels or image noise.

[0103] Furthermore, when testing the lightweight crack segmentation network using a second crack segmentation image with labeled information, as shown in Table 2, without fine-tuning the second crack image, the lightweight crack segmentation network still achieved 80.32% miIoU, 87.89% miDice, and 4.5% FPR (false positive rate) on the second crack segmentation image with labeled information. Its overall performance is significantly better than other segmentation network models. The lower FPR indicates that the lightweight crack segmentation network has strong resistance to false positives in complex environments.

[0104] Table 2 Performance Comparison Table

[0105]

[0106] In some embodiments, the deployment position of the multi-level edge detail enhancement module can also be obtained, and a lightweight crack segmentation network is constructed based on the embedding position. Specifically, the multi-level edge detail enhancement module is deployed to the full decoder, so that the edge feature extraction capability and detail restoration capability of the lightweight crack segmentation network can be improved.

[0107] In one embodiment, a plurality of crack models are established based on the second crack parameters, and boundary conditions and loads are applied to each crack model for simulation to obtain a plurality of second stress intensity factors, including:

[0108] The second crack parameters are sampled and combined to obtain a plurality of target crack parameter combinations.

[0109] A crack model is established based on each target crack parameter combination, respectively, and boundary conditions and loads are applied to each crack model to obtain an initial stress intensity factor corresponding to each target crack parameter combination;

[0110] The initial stress intensity factors are normalized to obtain a second stress intensity factor corresponding to each target crack parameter combination.

[0111] The sampling method includes but is not limited to Latin hypercube sampling and Monte Carlo sampling. Latin hypercube sampling is a statistical method for generating parameter combinations, which is particularly suitable for sampling in multi-dimensional space. Latin hypercube sampling aims to ensure that sample points are uniformly distributed in each dimension, so as to more effectively cover the entire input space. Compared with simple random sampling, Latin hypercube sampling can provide better space filling and fewer sample numbers to achieve similar accuracy.

[0112] Sampling and combination are sampling from different dimensional data and combining the second crack parameters sampled from different dimensions. Specifically, the value range of the second crack parameter is divided into several non-overlapping intervals in each dimension, so that the probability of each interval is the same; a value is randomly selected from each interval as a sample point, and it is ensured that the projections of these sample points in different dimensions are not repeated; the sample points from each dimension are combined to form a series of target crack parameter combinations. Further, since the target crack parameter combination is a combination of data in different dimensions, the target crack parameter combination can also be understood as the second crack parameter of the target object under various different working conditions. By sampling and combining, more simulation samples can be obtained, and more second stress intensity factors can be output, so as to improve the training effect of the regression model.

[0113] The crack model is established by a finite element simulation software. Each target crack parameter combination can be used to construct a crack model.

[0114] The normalized initial stress intensity factor can peel off the influence of load and size effect, thereby facilitating the training of the regression model. Since the target crack parameter combination can represent a crack working condition, the second stress intensity factor can also be understood as a stress intensity factor under a certain working condition.

[0115] In this embodiment, by sampling and combining the second crack parameters, a large number of simulation samples can be generated from a small number of second crack parameters, significantly improving the modeling efficiency and sample diversity. By establishing a crack model based on each target crack parameter combination respectively, and applying boundary conditions and loads to each crack model, various working conditions that the target object may exist can be simulated, thereby obtaining the initial stress intensity factors under various working conditions. By normalizing each initial stress intensity factor, the influence of load and size effect can be peeled off, facilitating the training of the regression model.

[0116] In some embodiments, the second stress intensity factors output by each crack model, the target crack parameter combinations, and the loads applied by each crack model can be stored in the same database, and since the second stress intensity factor is a stress intensity factor under a certain working condition, the database can also be understood as a large-scale crack mechanics database covering multiple working conditions. The second stress intensity factors stored in the database have a nonlinear mapping relationship with the target crack parameter combinations and the loads. For example, if the second stress intensity factor A is obtained through the target crack parameter combination a and the load x, and the second stress intensity factor B is obtained through the target crack parameter combination b and the load y, then the second stress intensity factor A has a nonlinear mapping relationship with the target crack parameter combination a and the load x, and the second stress intensity factor B has a nonlinear mapping relationship with the target crack parameter combination b and the load y. By constructing a database including the second stress intensity factors and the target crack parameter combinations, when training the regression model, data can be directly extracted from the database for training, without the need to reacquire the second stress intensity factors and the target crack parameter combinations, saving the time for model training.

[0117] In one embodiment, the regression model for predicting the stress intensity factor is trained based on the second crack parameters and the second stress intensity factors, including:

[0118] A plurality of regression models to be trained are determined.

[0119] Each regression model is trained based on the target crack parameter combination and the second stress intensity factor until the regression model converges, and the predicted value output by each regression model is obtained.

[0120] Based on the predicted value of each regression model and the second stress intensity factor, each regression model is evaluated to obtain an evaluation result of each regression model.

[0121] The regression model with the optimal evaluation result is determined as the regression model for predicting the stress intensity factor.

[0122] The plurality of regression models to be trained can be multiple types of linear regression, multiple types of polynomial regression, multiple types of ridge regression, multiple types of lasso regression, multiple types of elastic net regression, multiple types of logistic regression, multiple types of support vector regression, multiple types of decision tree regression, multiple types of random forest regression, multiple types of gradient boosting regression, K-Nearest Neighbors, Huber regression, and multiple types of partial least squares, or multiple types of other regression models.

[0123] Further, when training the regression model, the target crack parameter combination, the load, and the second stress intensity factor are input into the regression model to be trained, and the regression model is trained until the model converges. The target parameter combination and the load input into the regression model to be trained have a nonlinear mapping relationship with the second stress intensity factor. The essence of training the regression model is to establish a nonlinear mapping relationship from the second crack parameter and the load to the second stress intensity factor, so as to improve the prediction efficiency and generalization ability of the regression model.

[0124] The predicted value output by the regression model is also the stress intensity factor, but it is the stress intensity factor predicted based on the target crack parameter combination and the load, and the output predicted value is also a normalized value.

[0125] When evaluating the regression model, the regression model can be evaluated by at least one evaluation index to obtain an evaluation result of each regression model. The evaluation index can be calculated based on the predicted value output by the regression model and the second stress intensity factor corresponding to the predicted value. The second stress intensity factor corresponding to the predicted value is the second stress intensity factor corresponding to the target crack parameter combination used to obtain the predicted value. For example, the predicted value of the regression model for the target crack combination A is a, and the second stress intensity factor corresponding to the predicted value a is the second stress intensity factor corresponding to the target crack parameter combination A.

[0126] The regression model with the optimal evaluation result can be understood as the model with the best performance according to the selected evaluation standard or index.

[0127] In some embodiments, the regression model with the best evaluation result is one of CatBoost, XGBoost, random forest regression and LightGBM. Specifically, it is found in further analysis of the prediction distribution of each model on the test set that the prediction results of CatBoost, XGBoost, random forest regression and LightGBM are closely related to the true values, and the overall distribution of data points is close to the ideal fitting line y = x, indicating that these models can better capture the nonlinear characteristics in the data and have excellent fitting ability. In particular, CatBoost, whose predicted value almost completely fits the ideal line, has the best stability and accuracy. However, traditional regression models such as linear regression and support vector regression have a large deviation in a certain interval and fail to fully model the nonlinear impact of crack size and loading conditions on the stress intensity factor.

[0128] In this embodiment, each regression model is evaluated based on the predicted value of each regression model and the second stress intensity factor to obtain the evaluation result of each regression model, so that the best regression model can be effectively identified according to the evaluation result, and the regression model used for stress intensity factor prediction finally has the highest prediction accuracy and reliability.

[0129] In one embodiment, each regression model is evaluated based on the predicted value of each regression model and the second stress intensity factor to obtain the evaluation result of each regression model, including:

[0130] Based on the predicted value of each regression model and the second stress intensity factor, at least one of the determination coefficient, the mean absolute error and the mean square error of each regression model is calculated.

[0131] Based on at least one of the determination coefficient, the mean absolute error and the mean square error of each regression model, the evaluation result of each regression model is determined.

[0132] Wherein, the determination coefficient, the mean absolute error and the mean square error are evaluation indexes of the regression model. Further, if there are multiple evaluation indexes, each evaluation index is assigned a weight, the total score of each evaluation index after weighting is calculated, and the evaluation result of each regression model is determined according to the total score. When there are multiple evaluation indexes, a more comprehensive perspective can be provided for model evaluation, so that each regression model can be accurately evaluated.

[0133] The mean square error refers to the average of the square of the difference between the predicted value and the second stress intensity factor corresponding to the predicted value. The mean absolute error refers to the average of the absolute value of the difference between the predicted value and the second stress intensity factor corresponding to the predicted value. The coefficient of determination refers to the proportion of data variation explained by the regression model, and the coefficient of determination ranges from negative infinity to 1.

[0134] In this embodiment, at least one of the coefficient of determination, the mean absolute error and the mean square error of each regression model is calculated based on the predicted value and the second stress intensity factor of each regression model, and the evaluation result of each regression model is determined based on at least one of the coefficient of determination, the mean absolute error and the mean square error of each regression model, so that the regression model most suitable for the current second crack image and the application scenario can be selected, thereby improving the accuracy and reliability of the prediction.

[0135] The application also provides an application scenario applying the above-mentioned object state evaluation method. Specifically, the object state evaluation method is applied in the application scenario as follows:

[0136] The control unit constructs a lightweight crack segmentation network including an enhanced multi-scale attention fusion module, a multi-level edge detail enhancement module encoder, and a decoder. The structure diagram of the lightweight crack segmentation network is shown in FIG. 14. Specifically, the lightweight crack segmentation network includes an input layer, an encoder, feature processing, edge enhancement, a decoder, and an output layer. The input layer is used to input the second crack image and perform convolution processing through a double convolution layer to obtain a first feature map. In the double convolution layer, the second crack image is input through 3 channels, three feature maps are obtained through 3 convolution kernels, four feature maps are obtained through four convolution kernels, and a two-dimensional batch normalization layer, an activation function, and an efficient channel attention mechanism are combined to obtain the first feature map. In the efficient channel attention mechanism, a three-dimensional tensor X ∈ R C×H×W, C is the number of channels, H and W are the height and width of the first feature map respectively. In operation, a global average pooling operation is performed on each channel to compress the spatial dimension of each channel into a scalar value, and the kernel size is adaptively selected, for example, k = 5. The encoder includes down-sampling layer 1, down-sampling layer 2, down-sampling layer 3, and down-sampling layer 4, down-sampling layer 1 is used to down-sample the first feature map to obtain the second feature map, down-sampling layer 2 is used to down-sample the second feature map to obtain the third feature map, down-sampling layer 3 is used to down-sample the third feature map to obtain the fourth feature map, and down-sampling layer 4 is used to down-sample the fourth feature map to obtain the fifth feature map. The feature processing part includes enhanced multi-scale attention fusion module 1, enhanced multi-scale attention fusion module 2, and enhanced multi-scale attention fusion module 3, enhanced multi-scale attention fusion module 1 is used to fuse and process the fifth feature map and the fourth feature map to obtain the first fused feature, enhanced multi-scale attention fusion module 2 is used to fuse and process the first fused feature and the third feature map to obtain the second fused feature, and enhanced multi-scale attention fusion module 3 is used to fuse and process the second fused feature and the second feature map to obtain the third fused feature. The edge enhancement part includes multi-level edge detail enhancement module 1, multi-level edge detail enhancement module 2, multi-level edge detail enhancement module 3, and multi-level edge detail enhancement module 4, multi-level edge detail enhancement module 1 is used to perform edge enhancement processing on the fifth feature map to obtain the first up-sampling feature map, multi-level edge detail enhancement module 2 is used to perform edge enhancement processing on the first fused feature, multi-level edge detail enhancement module 3 is used to perform edge enhancement processing on the second fused feature, and multi-level edge detail enhancement module 4 is used to perform edge enhancement processing on the third fused feature. The decoder includes up-sampling layer 1, up-sampling layer 2, up-sampling layer 3, and up-sampling layer 4, up-sampling layer 1 is used to up-sample the first up-sampling feature map and the result of edge enhancement of the first fused feature to obtain the second up-sampling feature map, up-sampling layer 2 is used to up-sample the second up-sampling feature map and the result of edge enhancement of the second fused feature to obtain the third up-sampling feature map, up-sampling layer 3 is used to up-sample the third up-sampling feature map and the result of edge enhancement of the third fused feature to obtain the fourth up-sampling feature map, and up-sampling layer 4 is used to up-sample the fourth up-sampling feature map and the first feature map to obtain the fifth up-sampling feature map. In each up-sampling layer, up-sampling, splicing, and double convolution layers are combined. The output layer performs convolution processing on the fifth up-sampling feature map through the output convolution layer to obtain the segmentation result of the second crack image, so as to obtain the second crack parameter according to the segmentation result of the second crack image. The output convolution layer belongs to the down-sampling layer and is a combination of a two-dimensional maximum pooling layer and a double convolution layer.

[0137] The control unit obtains multiple second crack images, and inputs the obtained second crack images into the lightweight crack segmentation network to automatically perform pixel-level recognition to obtain a segmentation result. The control unit extracts coordinate information of the crack from the segmentation result, and converts the coordinate information using the conversion parameter to obtain converted crack coordinates when the conversion parameter is obtained. The control unit determines the second crack parameter based on the crack coordinates and outputs the second crack parameter. The control unit performs Latin hypercube sampling and combination based on the second crack parameter to obtain multiple target crack parameter combinations, and establishes a crack model based on the multiple target crack parameter combinations to perform finite element analysis to obtain an initial stress intensity factor corresponding to each target crack parameter combination. The control unit normalizes each initial stress intensity factor to obtain a second stress intensity factor corresponding to each target crack parameter combination.

[0138] The control unit trains each regression model based on the target crack parameter combination and the second stress intensity factor until each regression model converges to obtain a prediction value output by each regression model. The control unit calculates a determination coefficient, a mean absolute error and a mean square error of each regression model based on the prediction value of each regression model and the second stress intensity factor. The control unit determines an evaluation result of each regression model based on the determination coefficient, the mean absolute error and the mean square error of each regression model. The regression model with the optimal evaluation result is determined as the regression model for predicting the stress intensity factor. The specific flowchart is shown in Figure 15

[0139] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the direction of the arrow, these steps are not necessarily executed in sequence according to the direction of the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or stages in other steps.

[0140] Based on the same inventive concept, the embodiments of the present application also provide a state evaluation device of an object for implementing the state evaluation method of the object as described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, and therefore the specific limitations in one or more object state evaluation device embodiments provided below can refer to the limitations of the object state evaluation method described above, which will not be described here again.

[0141] In one embodiment, as Figure 16 ​As shown, a state evaluation device of an object is provided, comprising:

[0142] An image acquisition unit 1602 is configured to acquire a first crack image of an object to be evaluated;

[0143] A prediction unit 1604 is configured to use a regression model to predict a first stress intensity factor of the object to be evaluated based on a first crack parameter in the first crack image; the regression model is a model for predicting a stress intensity factor, and the regression model is obtained by training a second crack parameter in a second crack image of a target object and a second stress intensity factor simulated based on the second crack parameter.

[0144] A state evaluation unit 1606 is configured to evaluate the state of the object to be evaluated based on the first stress intensity factor, and obtain a state evaluation result of the object to be evaluated.

[0145] In one of the embodiments, the state evaluation device of the object is further configured to: acquire a second crack parameter in a second crack image of a target object; establish a plurality of crack models based on the second crack parameter, and apply boundary conditions and loads to each crack model for simulation to obtain a plurality of second stress intensity factors; and train a regression model based on the second crack parameter and the second stress intensity factors to obtain the regression model for predicting a stress intensity factor.

[0146] In one of the embodiments, the state evaluation device of the object is further configured to: acquire coordinate information of a crack in a second crack image of a target object; when the conversion parameter is acquired, convert the coordinate information using the conversion parameter to obtain converted crack coordinates; and determine the second crack parameter based on the crack coordinates.

[0147] In one of the embodiments, the state evaluation device of the object is further configured to: establish a lightweight crack segmentation network; the lightweight crack segmentation network comprises an enhanced multi-scale attention fusion module and a multi-level edge detail enhancement module; and acquire the coordinate information of the crack in the second crack image of the target object using the lightweight crack segmentation network.

[0148] In one of the embodiments, the state evaluation device of the object is further configured to: sample and combine the second crack parameter to obtain a plurality of target crack parameter combinations; establish a crack model based on each target crack parameter combination, respectively, and apply boundary conditions and loads to each crack model to obtain an initial stress intensity factor corresponding to each target crack parameter combination; and perform normalization processing on the initial stress intensity factors to obtain a second stress intensity factor corresponding to each target crack parameter combination.

[0149] In one embodiment, the object state assessment device is further configured to: determine multiple regression models to be trained; train each regression model based on the target crack parameter combination and the second stress intensity factor until each regression model converges, and obtain the predicted value output by each regression model; evaluate each regression model based on the predicted value of each regression model and the second stress intensity factor, and obtain the evaluation result of each regression model; and determine the regression model with the best evaluation result as the regression model for predicting the stress intensity factor.

[0150] In one embodiment, the object state assessment device is further configured to: calculate at least one of the determination coefficient, mean absolute error, and mean square error of each regression model based on the predicted value of each regression model and the second stress intensity factor; and determine the assessment result of each regression model based on at least one of the determination coefficient, mean absolute error, and mean square error of each regression model.

[0151] Each module in the aforementioned object status assessment device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0152] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 17 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores a first crack image, first crack parameters, a first stress intensity factor, state assessment results, second crack parameters, a second stress intensity factor, coordinate information, crack coordinates, a target crack parameter combination, an initial stress intensity factor, and the assessment results of a regression model. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a state assessment method for an object.

[0153] Those skilled in the art will understand that Figure 17 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0154] In an embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above method embodiments when executing the computer program.

[0155] In an embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program implementing the steps in the above method embodiments when executed by a processor.

[0156] In an embodiment, a computer program product is provided, including a computer program, and the computer program implementing the steps in the above method embodiments when executed by a processor.

[0157] In an embodiment, a state evaluation system of an object is provided, including a crack detection unit, a parameter extraction and calibration unit, a simulation unit, a training and prediction unit, and a control unit; the crack detection unit is configured to detect cracks in a first crack image of an object to be evaluated and cracks in a second crack image of a target object; the parameter conversion unit is configured to obtain coordinate information of the cracks in the second crack image and convert the coordinate information; the simulation unit is configured to obtain a second stress intensity factor of the target object; the training and prediction unit is configured to train a regression model and predict the stress intensity factor; and the control unit is configured to implement the steps in the above method embodiments.

[0158] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0159] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric memory (Ferroelectric Random Access Memory, FRAM), phase change memory (Phase Change Memory, PCM), graphene memory, etc. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0160] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0161] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method of state assessment of an object, characterized by, The method comprises: acquiring a first crack image of an object to be evaluated; based on the first crack parameter in the first crack image, using a regression model for prediction to obtain the first stress intensity factor of the object to be evaluated; the regression model is a model for predicting the stress intensity factor, the regression model is trained by the second crack parameter and the second stress intensity factor in the second crack image of the target object, the second stress intensity factor is simulated by the second crack parameter; the second crack parameter includes crack length, size ratio, bifurcation angle and stress ratio; based on the first stress intensity factor, the state of the object to be evaluated is evaluated to obtain the state evaluation result of the object to be evaluated; The method for acquiring the coordinate information of the crack in the second crack image of the target object comprises: establishing a lightweight crack segmentation network; the lightweight crack segmentation network comprises an enhanced multi-scale attention fusion module, a multi-level edge detail enhancement module, a decoder and an encoder, the decoder comprises a plurality of up-sampling layers, the encoder comprises a plurality of down-sampling layers, each down-sampling layer is connected with an enhanced multi-scale attention fusion module, each enhanced multi-scale attention fusion module is connected with a multi-level edge detail enhancement module, and each multi-level edge detail enhancement module is connected with an up-sampling layer; the multi-level edge detail enhancement module is used for performing edge enhancement processing on the fusion features output by the enhanced multi-scale attention fusion module, and the up-sampling layer is used for performing up-sampling on the edge enhancement results output by the multi-level edge detail enhancement module; the coordinate information of the crack in the second crack image of the target object is acquired by using the lightweight crack segmentation network; the coordinate information is used for determining the second crack parameter in the second crack image.

2. The method of claim 1, wherein, The method further comprises: acquiring the second crack parameter in the second crack image of the target object; based on the second crack parameter, a plurality of crack models are established, and boundary conditions and loads are applied to each crack model for simulation to obtain a plurality of second stress intensity factors; based on the second crack parameter and each second stress intensity factor, a regression model is trained to obtain a regression model for predicting the stress intensity factor.

3. The method of claim 2, wherein, The method for acquiring the second crack parameter in the second crack image of the target object comprises: acquiring the coordinate information of the crack in the second crack image of the target object; when the conversion parameter is acquired, the coordinate information is converted by using the conversion parameter to obtain the converted crack coordinate; based on the crack coordinate, the second crack parameter is determined.

4. The method of claim 2, wherein, The method for acquiring the second crack parameter in the second crack image of the target object comprises: sampling and combining the second crack parameter to obtain a plurality of target crack parameter combinations; based on each target crack parameter combination, a crack model is established, and boundary conditions and loads are applied to each crack model to obtain an initial stress intensity factor corresponding to each target crack parameter combination; The initial stress intensity factors are normalized to obtain a second stress intensity factor corresponding to each target crack parameter combination.

5. The method of claim 4, wherein, The regression model is trained based on the second crack parameters and the second stress intensity factors to obtain a regression model for predicting stress intensity factors, including: determining a plurality of regression models to be trained; training each regression model based on the target crack parameter combination and the second stress intensity factor until each regression model converges to obtain a predicted value output by each regression model; evaluating each regression model based on the predicted value of each regression model and the second stress intensity factor to obtain an evaluation result of each regression model; determining the regression model with the optimal evaluation result as the regression model for predicting stress intensity factors.

6. The method of claim 5, wherein, The evaluation result of each regression model is obtained by evaluating each regression model based on the predicted value of each regression model and the second stress intensity factor, including: calculating at least one of a determination coefficient, a mean absolute error, and a mean square error of each regression model based on the predicted value of each regression model and the second stress intensity factor; determining the evaluation result of each regression model based on at least one of the determination coefficient, the mean absolute error, and the mean square error of each regression model.

7. A state evaluation device of an object, characterized by, The device comprises: an image acquisition unit configured to acquire a first crack image of an object to be evaluated; a prediction unit configured to use a regression model to predict a first stress intensity factor of the object to be evaluated based on first crack parameters in the first crack image, wherein the regression model is a model for predicting stress intensity factors, the regression model is trained by second crack parameters in a second crack image of a target object and a second stress intensity factor, the second stress intensity factor is simulated based on the second crack parameters, the second crack parameters include crack length, size ratio, bifurcation angle, and stress ratio, and the method for obtaining coordinate information of a crack in the second crack image of the target object comprises: establishing a lightweight crack segmentation network, wherein the lightweight crack segmentation network comprises an enhanced multi-scale attention fusion module, a multi-level edge detail enhancement module, a decoder, and an encoder, the decoder comprises a plurality of up-sampling layers, the encoder comprises a plurality of down-sampling layers, each down-sampling layer is connected to one enhanced multi-scale attention fusion module, each enhanced multi-scale attention fusion module is connected to one multi-level edge detail enhancement module, each multi-level edge detail enhancement module is connected to one up-sampling layer, the multi-level edge detail enhancement module is configured to perform edge enhancement processing on fusion features output by the enhanced multi-scale attention fusion module, and the up-sampling layer is configured to perform up-sampling on edge enhancement results output by the multi-level edge detail enhancement module, and the coordinate information of the crack in the second crack image of the target object is obtained using the lightweight crack segmentation network, and the coordinate information is used to determine the second crack parameters in the second crack image. The state evaluation unit is configured to evaluate a state of the object to be evaluated based on the first stress intensity factor, and obtain a state evaluation result of the object to be evaluated. 8.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-7. The processor, when executing the computer program, implements the steps of the method in any one of claims 1 to 6. 9.A state evaluation system of an object, comprising a crack detection unit, a parameter conversion unit, a simulation unit, a training and prediction unit, and a control unit, and the state evaluation system is characterized in that: The crack detection unit is configured to detect cracks in a first crack image of an object to be evaluated, and detect cracks in a second crack image of a target object; The parameter conversion unit is configured to obtain coordinate information of the cracks in the second crack image, and convert the coordinate information; The simulation unit is configured to obtain a second stress intensity factor of the target object; The training and prediction unit is configured to train a regression model, and predict the stress intensity factor; The control unit is configured to implement the steps of the method in any one of claims 1 to 6.

Images